Consent without Disclosure: Trade Secrets, Copyright and the Black Box of AI Training Data in India

Abstract

When a Bengaluru home-services start-up, Pronto, was reported in May 2026 to be using in-home cameras during service bookings, and declined to say whether the resulting footage would feed India's emerging ‘physical AI’ and robotics industry, it exposed a gap that copyright and privacy scholarship in India have so far examined separately: the space between ‘consent to be recorded’ and ‘consent to become training data’. This essay argues that trade secrecy — which AI developers routinely invoke to withhold details of the origin, composition and downstream use of training datasets — functions as a shield that can render both the Digital Personal Data Protection Act 2023's consent architecture and the Copyright Act 1957's infringement remedies practically unenforceable. Drawing on comparative material from the European Union's AI Act and on recent litigation in the United States, the essay develops a four-part ‘Legality-Relevance Test’ and a five-layer disclosure model suited to Indian civil procedure.